Improved Kernel based Tracking for Fast Moving Object

A novel approach of discriminative object representation and multiple-kernel tracking is proposed. We first employ a discriminative object representation, which introduces the foreground and background modelling ingredient to select the most discriminative features from a set of candidates via classification procedure. In the context of using kernel based tracking algorithm, a multiple-kernel strategy is employed to handle the difficulties resulted from fast motion through refining the ill-initialization position according to prerefinement method. Extensive experiments demonstrate that the proposed tracker works better than Camshift and traditional kernel tracker.